auto-unwrap: make `foo()!` the same as `foo().unwrap()`

[liigo]

Someone want to remove one of .foo() and .foo_opt(), but that will let us write more foo().unwrap(), which is a bit nasty. I'd like to see introducing a new syntax sugar: foo()!, which means the same as foo().unwrap(). The ! here means: 1) ensure it's not None, 2) it maybe fail!.

We can define a trait Unwrapable<T> that has an fn unwrap() -> T, and impl it for Option, Result<T,E> and others. When ! operated on an Unwrapable value, rustc call its unwrap() automatically. (I do not know whether or not it will be ambiguous with macro!.)

After it is accepted, we may safely remove one of .foo() and .foo_opt().

[jfager]

Isn't the reason having 'unwrap' everywhere is a smell is because it ignores the safety net Option gives you and risks task failure everywhere it's used? This does the exact same thing but seems to bless it with special syntax.

At least when you have to type out the full 'unwrap()' method you're having to work a little bit for it, and it's clear from it's ugliness that it should maybe be avoided. From an affordance standpoint, it seems unfortunate that the 'wrong' thing (just fail) is easier to reach for than the 'right' thing (generate a default or otherwise handle the error), so maybe it should even be a little uglier, and all Option methods that can fail should be removed, forcing you to write out let a = foo.unwrap_or_else(|| fail()!) every time.

If there's going to be special handling for Option or other Unwrappables, it should take the form of something like do-notation or Option chaining.

[liigo]

@jfager There are many .unwrap() in rust source repository. Are they all misusage? I don't think so. This is a common usage, authors need it, they ensure got Some(T) not None in that situation, and even fail! is acceptable. Are you like code full of let a = foo.unwrap_or_else(|| fail()!) everywhere? match need at least 4 lines.

[aturon]

cc me

[pnkfelix]

Closing: A change like this needs to go through the RFC process that we imposed earlier this year: https://github.com/rust-lang/rfcs/